"The 'it' in AI models is the dataset" resurfaces, fueling frontier lab data debate
teortaxesTex · x · 2026-11-21
teortaxesTex pushes back on LinusMixson's angry take that Chinese labs only produced "braindamaged copies of Claude" via distillation, arguing Thinking Machines Lab is a genuine frontier lab whose people "have been in that Slack" and know the vendors.
The attached screenshot is a 2023 essay by an OpenAI insider, The "it" in AI models is the dataset, arguing that models trained long enough on the same dataset converge to the same point regardless of architecture — diffusion and AR sampling produce the same images — so model behavior is determined by the dataset, nothing else. The exchange rekindles the data-vs-architecture debate and questions about where frontier labs get their data.
More from Companies & People
- Listener slams Amodei siblings' Isaacson interview as elitist, contrasting Dario's humble 2023 Lex appearance — GaryMarcus · 2026-11-21
- Report: Nearly 5 OpenAI Staff Left in 14 Hours Amid Safety Policy Shake-up — GaryMarcus · 2026-11-21
- Viral Jensen Huang clip misread as insult—context shows he was praising Intel exec's intellect — kevinnbass · 2026-11-21
- Thiel doubts AI lab monopolies, xAI rebuild began when Musk showed up, and more signals — himanshustwts · 2026-11-21
- Palo Alto Networks won't close an acquisition until the founder signs a joint product roadmap — himanshustwts · 2026-11-21
- Stripe operating lessons: changing strategy feels like progress but usually isn't — intellectronica · 2026-11-21